Presenter: Federico Cerutti
Wednesday, June 21st, 2023 11:00
Location: Sala Maxwell, 5th floor Corso Castelfidardo 34
When collaborating with an AI system, we must assess when to trust its recommendations. Suppose we mistakenly trust it in regions where it is likely to err. In that case, catastrophic failures may occur, hence the need for Bayesian approaches for reasoning and learning to determine the confidence (or epistemic uncertainty) in the probabilities of the queried outcome. Pure Bayesian methods, however, suffer from high computational costs. To overcome them, we revert to efficient and effective approximations. This talk will discuss some techniques that take the name of evidential reasoning and learning, from the Bayesian update of given hypotheses based on additional evidence collected. We will discuss probabilistic circuits with uncertain probabilities, uncertainty-aware deep classifiers, and current research directions in neuro-symbolic settings, including a recent application to passive radars.
Federico Cerutti is an Associate Professor at the University of Brescia, Italy, lecturing on Digital Transformation and Computer Security. He is the Chair of the local branch (node) of the Italian Cybersecurity National Laboratory and recently served as a Group of Expert Evaluators member for the 2015-2019 edition of the Italian Evaluation of Research Quality (VQR) exercise.
He is also an Honorary Senior Lecturer at Cardiff University (UK) and a Visiting Fellow at the University of Southampton (UK).
His research focuses on learning and reasoning with uncertain and sparse data for supporting situational understanding and (cyber-threat) intelligence analysis.